muzerai/kanana-nano-2.1b-instruct-creative-de-aijoa
Abliterated vs kakaocorp/kanana-nano-2.1b-instruct. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 32 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 0/96, Attention 32/128
1 derivatives checked
- mradermacher/kanana-nano-2.1b-instruct-creative-de-aijoa-GGUF: Same weights, converted